Parallel Algorithm and Processor Selection Based on Fuzzy Logic

The face of parallel computing has changed in the last few years as high performance clusters of workstations are being used in conjunction with supercomputers to solve demanding computational problems. In order for a user to effectively run an application on both tightly coupled and network based clusters, he must often use different algorithms that are suited to the network available on the computing platform. An application may also be able to effectively utilize a different number of processing nodes with a particular algorithm and processor configuration. It is difficult for a user to determine which set of parameters to select in order to customize the application for an available computing environment. The principal aim of this research is to show that fuzzy logic can be used to select the most efficient algorithm and an optimal number of processors for a parallel application. In this paper we examine three algorithms for image convolution which each have advantages depending on the available architecture and problem size. A fuzzy logic technique is developed which is able to make effective selections, freeing the user from an otherwise daunting task. The fuzzy logic selection system is easy to set up and these results can be extended to additional applications.